hive-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Mostafa Mokhtar (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (HIVE-8261) CBO : Predicate pushdown is removed by Optiq
Date Thu, 25 Sep 2014 22:51:36 GMT

    [ https://issues.apache.org/jira/browse/HIVE-8261?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14148439#comment-14148439
] 

Mostafa Mokhtar commented on HIVE-8261:
---------------------------------------

[~julianhyde]

> CBO : Predicate pushdown is removed by Optiq 
> ---------------------------------------------
>
>                 Key: HIVE-8261
>                 URL: https://issues.apache.org/jira/browse/HIVE-8261
>             Project: Hive
>          Issue Type: Bug
>          Components: CBO
>    Affects Versions: 0.14.0, 0.13.1
>            Reporter: Mostafa Mokhtar
>            Assignee: Gunther Hagleitner
>             Fix For: 0.14.0
>
>
> Plan for TPC-DS Q64 wasn't optimal upon looking at the logical plan I realized that predicate
pushdown is not applied on date_dim d1.
> Interestingly before optiq we have the predicate pushed :
> {code}
> HiveFilterRel(condition=[<=($5, $1)])
>     HiveJoinRel(condition=[=($3, $6)], joinType=[inner])
>       HiveProjectRel(_o__col0=[$0], _o__col1=[$2], _o__col2=[$3], _o__col3=[$1])
>         HiveFilterRel(condition=[=($0, 2000)])
>           HiveAggregateRel(group=[{0, 1}], agg#0=[count()], agg#1=[sum($2)])
>             HiveProjectRel($f0=[$4], $f1=[$5], $f2=[$2])
>               HiveJoinRel(condition=[=($1, $8)], joinType=[inner])
>                 HiveJoinRel(condition=[=($1, $5)], joinType=[inner])
>                   HiveJoinRel(condition=[=($0, $3)], joinType=[inner])
>                     HiveProjectRel(ss_sold_date_sk=[$0], ss_item_sk=[$2], ss_wholesale_cost=[$11])
>                       HiveTableScanRel(table=[[tpcds_bin_partitioned_orc_200.store_sales]])
>                     HiveProjectRel(d_date_sk=[$0], d_year=[$6])
>                       HiveTableScanRel(table=[[tpcds_bin_partitioned_orc_200.date_dim]])
>                   HiveFilterRel(condition=[AND(in($2, 'maroon', 'burnished', 'dim', 'steel',
'navajo', 'chocolate'), between(false, $1, 35, +(35, 10)), between(false, $1, +(35, 1), +(35,
15)))])
>                     HiveProjectRel(i_item_sk=[$0], i_current_price=[$5], i_color=[$17])
>                       HiveTableScanRel(table=[[tpcds_bin_partitioned_orc_200.item]])
>                 HiveProjectRel(_o__col0=[$0])
>                   HiveAggregateRel(group=[{0}])
>                     HiveProjectRel($f0=[$0])
>                       HiveJoinRel(condition=[AND(=($0, $2), =($1, $3))], joinType=[inner])
>                         HiveProjectRel(cs_item_sk=[$15], cs_order_number=[$17])
>                           HiveTableScanRel(table=[[tpcds_bin_partitioned_orc_200.catalog_sales]])
>                         HiveProjectRel(cr_item_sk=[$2], cr_order_number=[$16])
>                           HiveTableScanRel(table=[[tpcds_bin_partitioned_orc_200.catalog_returns]])
>       HiveProjectRel(_o__col0=[$0], _o__col1=[$2], _o__col3=[$1])
>         HiveFilterRel(condition=[=($0, +(2000, 1))])
>           HiveAggregateRel(group=[{0, 1}], agg#0=[count()])
>             HiveProjectRel($f0=[$4], $f1=[$5], $f2=[$2])
>               HiveJoinRel(condition=[=($1, $8)], joinType=[inner])
>                 HiveJoinRel(condition=[=($1, $5)], joinType=[inner])
>                   HiveJoinRel(condition=[=($0, $3)], joinType=[inner])
>                     HiveProjectRel(ss_sold_date_sk=[$0], ss_item_sk=[$2], ss_wholesale_cost=[$11])
>                       HiveTableScanRel(table=[[tpcds_bin_partitioned_orc_200.store_sales]])
>                     HiveProjectRel(d_date_sk=[$0], d_year=[$6])
>                       HiveTableScanRel(table=[[tpcds_bin_partitioned_orc_200.date_dim]])
>                   HiveFilterRel(condition=[AND(in($2, 'maroon', 'burnished', 'dim', 'steel',
'navajo', 'chocolate'), between(false, $1, 35, +(35, 10)), between(false, $1, +(35, 1), +(35,
15)))])
>                     HiveProjectRel(i_item_sk=[$0], i_current_price=[$5], i_color=[$17])
>                       HiveTableScanRel(table=[[tpcds_bin_partitioned_orc_200.item]])
>                 HiveProjectRel(_o__col0=[$0])
>                   HiveAggregateRel(group=[{0}])
>                     HiveProjectRel($f0=[$0])
>                       HiveJoinRel(condition=[AND(=($0, $2), =($1, $3))], joinType=[inner])
>                         HiveProjectRel(cs_item_sk=[$15], cs_order_number=[$17])
>                           HiveTableScanRel(table=[[tpcds_bin_partitioned_orc_200.catalog_sales]])
>                         HiveProjectRel(cr_item_sk=[$2], cr_order_number=[$16])
>                           HiveTableScanRel(table=[[tpcds_bin_partitioned_orc_200.catalog_returns]])
> {code}
> While after Optiq the filter on date_dim gets pulled up the plan 
> {code}
>   HiveFilterRel(condition=[<=($5, $1)]): rowcount = 1.0, cumulative cost = {5.50188454E8
rows, 0.0 cpu, 0.0 io}, id = 6895
>     HiveProjectRel(_o__col0=[$0], _o__col1=[$1], _o__col2=[$2], _o__col3=[$3], _o__col00=[$4],
_o__col10=[$5], _o__col30=[$6]): rowcount = 1.0, cumulative cost = {5.50188454E8 rows, 0.0
cpu, 0.0 io}, id = 7046
>       HiveJoinRel(condition=[=($3, $6)], joinType=[inner]): rowcount = 1.0, cumulative
cost = {5.50188454E8 rows, 0.0 cpu, 0.0 io}, id = 7041
>         HiveProjectRel(_o__col0=[$0], _o__col1=[$2], _o__col2=[$3], _o__col3=[$1]): rowcount
= 1.0, cumulative cost = {5.50188452E8 rows, 0.0 cpu, 0.0 io}, id = 6857
>           HiveFilterRel(condition=[=($0, 2000)]): rowcount = 1.0, cumulative cost = {5.50188452E8
rows, 0.0 cpu, 0.0 io}, id = 6855
>             HiveAggregateRel(group=[{0, 1}], agg#0=[count()], agg#1=[sum($2)]): rowcount
= 1.0, cumulative cost = {5.50188452E8 rows, 0.0 cpu, 0.0 io}, id = 6853
>               HiveProjectRel($f0=[$4], $f1=[$5], $f2=[$2]): rowcount = 1.0, cumulative
cost = {5.50188452E8 rows, 0.0 cpu, 0.0 io}, id = 6851
>                 HiveProjectRel(ss_sold_date_sk=[$3], ss_item_sk=[$4], ss_wholesale_cost=[$5],
d_date_sk=[$0], d_year=[$1], i_item_sk=[$6], i_current_price=[$7], i_color=[$8], _o__col0=[$2]):
rowcount = 1.0, cumulative cost = {5.50188452E8 rows, 0.0 cpu, 0.0 io}, id = 7039
>                   HiveJoinRel(condition=[=($3, $0)], joinType=[inner]): rowcount = 1.0,
cumulative cost = {5.50188452E8 rows, 0.0 cpu, 0.0 io}, id = 7037
>                     HiveProjectRel(d_date_sk=[$0], d_year=[$6]): rowcount = 73049.0,
cumulative cost = {0.0 rows, 0.0 cpu, 0.0 io}, id = 6861
>                       HiveTableScanRel(table=[[tpcds_bin_partitioned_orc_200.date_dim]]):
rowcount = 73049.0, cumulative cost = {0}, id = 6537
>                     HiveJoinRel(condition=[=($2, $0)], joinType=[inner]): rowcount =
1.0, cumulative cost = {5.50115402E8 rows, 0.0 cpu, 0.0 io}, id = 7035
>                       HiveProjectRel(_o__col0=[$0]): rowcount = 38846.0, cumulative cost
= {3.15348608E8 rows, 0.0 cpu, 0.0 io}, id = 6847
>                         HiveAggregateRel(group=[{0}]): rowcount = 38846.0, cumulative
cost = {3.15348608E8 rows, 0.0 cpu, 0.0 io}, id = 6845
>                           HiveProjectRel($f0=[$0]): rowcount = 6.692553251460564E8, cumulative
cost = {3.15348608E8 rows, 0.0 cpu, 0.0 io}, id = 6843
>                             HiveProjectRel(cs_item_sk=[$0], cs_order_number=[$1], cr_item_sk=[$2],
cr_order_number=[$3]): rowcount = 6.692553251460564E8, cumulative cost = {3.15348608E8 rows,
0.0 cpu, 0.0 io}, id = 6945
>                               HiveJoinRel(condition=[AND(=($0, $2), =($1, $3))], joinType=[inner]):
rowcount = 6.692553251460564E8, cumulative cost = {3.15348608E8 rows, 0.0 cpu, 0.0 io}, id
= 6940
>                                 HiveProjectRel(cs_item_sk=[$15], cs_order_number=[$17]):
rowcount = 2.86549727E8, cumulative cost = {0.0 rows, 0.0 cpu, 0.0 io}, id = 6871
>                                   HiveTableScanRel(table=[[tpcds_bin_partitioned_orc_200.catalog_sales]]):
rowcount = 2.86549727E8, cumulative cost = {0}, id = 6531
>                                 HiveProjectRel(cr_item_sk=[$2], cr_order_number=[$16]):
rowcount = 2.8798881E7, cumulative cost = {0.0 rows, 0.0 cpu, 0.0 io}, id = 6873
>                                   HiveTableScanRel(table=[[tpcds_bin_partitioned_orc_200.catalog_returns]]):
rowcount = 2.8798881E7, cumulative cost = {0}, id = 6532
>                       HiveJoinRel(condition=[=($1, $3)], joinType=[inner]): rowcount
= 1.0, cumulative cost = {5.50076555E8 rows, 0.0 cpu, 0.0 io}, id = 6996
>                         HiveProjectRel(ss_sold_date_sk=[$0], ss_item_sk=[$2], ss_wholesale_cost=[$11]):
rowcount = 5.50076554E8, cumulative cost = {0.0 rows, 0.0 cpu, 0.0 io}, id = 6859
>                           HiveTableScanRel(table=[[tpcds_bin_partitioned_orc_200.store_sales]]):
rowcount = 5.50076554E8, cumulative cost = {0}, id = 6538
>                         HiveFilterRel(condition=[AND(in($2, 'maroon', 'burnished', 'dim',
'steel', 'navajo', 'chocolate'), between(false, $1, 35, +(35, 10)), between(false, $1, +(35,
1), +(35, 15)))]): rowcount = 1.0, cumulative cost = {0.0 rows, 0.0 cpu, 0.0 io}, id = 6833
>                           HiveProjectRel(i_item_sk=[$0], i_current_price=[$5], i_color=[$17]):
rowcount = 48000.0, cumulative cost = {0.0 rows, 0.0 cpu, 0.0 io}, id = 6831
>                             HiveTableScanRel(table=[[tpcds_bin_partitioned_orc_200.item]]):
rowcount = 48000.0, cumulative cost = {0}, id = 6539
>         HiveProjectRel(_o__col0=[$0], _o__col1=[$2], _o__col3=[$1]): rowcount = 1.0,
cumulative cost = {5.50188452E8 rows, 0.0 cpu, 0.0 io}, id = 6891
>           HiveFilterRel(condition=[=($0, +(2000, 1))]): rowcount = 1.0, cumulative cost
= {5.50188452E8 rows, 0.0 cpu, 0.0 io}, id = 6889
>             HiveAggregateRel(group=[{0, 1}], agg#0=[count()]): rowcount = 1.0, cumulative
cost = {5.50188452E8 rows, 0.0 cpu, 0.0 io}, id = 6887
>               HiveProjectRel($f0=[$4], $f1=[$5], $f2=[$2]): rowcount = 1.0, cumulative
cost = {5.50188452E8 rows, 0.0 cpu, 0.0 io}, id = 6885
>                 HiveProjectRel(ss_sold_date_sk=[$3], ss_item_sk=[$4], ss_wholesale_cost=[$5],
d_date_sk=[$0], d_year=[$1], i_item_sk=[$6], i_current_price=[$7], i_color=[$8], _o__col0=[$2]):
rowcount = 1.0, cumulative cost = {5.50188452E8 rows, 0.0 cpu, 0.0 io}, id = 6992
>                   HiveJoinRel(condition=[=($3, $0)], joinType=[inner]): rowcount = 1.0,
cumulative cost = {5.50188452E8 rows, 0.0 cpu, 0.0 io}, id = 6990
>                     HiveProjectRel(d_date_sk=[$0], d_year=[$6]): rowcount = 73049.0,
cumulative cost = {0.0 rows, 0.0 cpu, 0.0 io}, id = 6861
>                       HiveTableScanRel(table=[[tpcds_bin_partitioned_orc_200.date_dim]]):
rowcount = 73049.0, cumulative cost = {0}, id = 6537
>                     HiveJoinRel(condition=[=($2, $0)], joinType=[inner]): rowcount =
1.0, cumulative cost = {5.50115402E8 rows, 0.0 cpu, 0.0 io}, id = 6988
>                       HiveProjectRel(_o__col0=[$0]): rowcount = 38846.0, cumulative cost
= {3.15348608E8 rows, 0.0 cpu, 0.0 io}, id = 6881
>                         HiveAggregateRel(group=[{0}]): rowcount = 38846.0, cumulative
cost = {3.15348608E8 rows, 0.0 cpu, 0.0 io}, id = 6879
>                           HiveProjectRel($f0=[$0]): rowcount = 6.692553251460564E8, cumulative
cost = {3.15348608E8 rows, 0.0 cpu, 0.0 io}, id = 6877
>                             HiveProjectRel(cs_item_sk=[$0], cs_order_number=[$1], cr_item_sk=[$2],
cr_order_number=[$3]): rowcount = 6.692553251460564E8, cumulative cost = {3.15348608E8 rows,
0.0 cpu, 0.0 io}, id = 6938
>                               HiveJoinRel(condition=[AND(=($0, $2), =($1, $3))], joinType=[inner]):
rowcount = 6.692553251460564E8, cumulative cost = {3.15348608E8 rows, 0.0 cpu, 0.0 io}, id
= 6933
>                                 HiveProjectRel(cs_item_sk=[$15], cs_order_number=[$17]):
rowcount = 2.86549727E8, cumulative cost = {0.0 rows, 0.0 cpu, 0.0 io}, id = 6871
>                                   HiveTableScanRel(table=[[tpcds_bin_partitioned_orc_200.catalog_sales]]):
rowcount = 2.86549727E8, cumulative cost = {0}, id = 6531
>                                 HiveProjectRel(cr_item_sk=[$2], cr_order_number=[$16]):
rowcount = 2.8798881E7, cumulative cost = {0.0 rows, 0.0 cpu, 0.0 io}, id = 6873
>                                   HiveTableScanRel(table=[[tpcds_bin_partitioned_orc_200.catalog_returns]]):
rowcount = 2.8798881E7, cumulative cost = {0}, id = 6532
>                       HiveJoinRel(condition=[=($1, $3)], joinType=[inner]): rowcount
= 1.0, cumulative cost = {5.50076555E8 rows, 0.0 cpu, 0.0 io}, id = 6949
>                         HiveProjectRel(ss_sold_date_sk=[$0], ss_item_sk=[$2], ss_wholesale_cost=[$11]):
rowcount = 5.50076554E8, cumulative cost = {0.0 rows, 0.0 cpu, 0.0 io}, id = 6859
>                           HiveTableScanRel(table=[[tpcds_bin_partitioned_orc_200.store_sales]]):
rowcount = 5.50076554E8, cumulative cost = {0}, id = 6538
>                         HiveFilterRel(condition=[AND(in($2, 'maroon', 'burnished', 'dim',
'steel', 'navajo', 'chocolate'), between(false, $1, 35, +(35, 10)), between(false, $1, +(35,
1), +(35, 15)))]): rowcount = 1.0, cumulative cost = {0.0 rows, 0.0 cpu, 0.0 io}, id = 6867
>                           HiveProjectRel(i_item_sk=[$0], i_current_price=[$5], i_color=[$17]):
rowcount = 48000.0, cumulative cost = {0.0 rows, 0.0 cpu, 0.0 io}, id = 6865
>                             HiveTableScanRel(table=[[tpcds_bin_partitioned_orc_200.item]]):
rowcount = 48000.0, cumulative cost = {0}, id = 6539
> {code}
> I simplified the query a little bit while still maintaining the query structure 
> The query : 
> Note that the final join between cs1 and cs2 has a predicates  "cs1.syear = 2000 and
cs2.syear = 2000 + 1"
> {code}
> select cs1.syear ,cs1.cnt
>      ,cs1.s1 ,cs2.syear ,cs2.cnt
> from
> (select d1.d_year as syear ,count(*) as cnt,sum(ss_wholesale_cost) as s1 ,i_item_sk as
item_sk
>   FROM   store_sales
>         JOIN date_dim d1 ON store_sales.ss_sold_date_sk = d1.d_date_sk
>         JOIN item ON store_sales.ss_item_sk = item.i_item_sk
>         JOIN
>  (select cs_item_sk
>   from catalog_sales JOIN catalog_returns
>   ON catalog_sales.cs_item_sk = catalog_returns.cr_item_sk
>     and catalog_sales.cs_order_number = catalog_returns.cr_order_number
>   group by cs_item_sk) cs_ui
> ON store_sales.ss_item_sk = cs_ui.cs_item_sk
>   WHERE  
>          i_color in ('maroon','burnished','dim','steel','navajo','chocolate') and
>          i_current_price between 35 and 35 + 10 and
>          i_current_price between 35 + 1 and 35 + 15
> group by d1.d_year,i_item_sk
> ) cs1
> JOIN
> (select d1.d_year as syear ,count(*) as cnt,sum(ss_wholesale_cost) as s1 , i_item_sk
as item_sk
>   FROM   store_sales
>         JOIN date_dim d1 ON store_sales.ss_sold_date_sk = d1.d_date_sk
>         JOIN item ON store_sales.ss_item_sk = item.i_item_sk
>         JOIN
>  (select cs_item_sk
>   from catalog_sales JOIN catalog_returns
>   ON catalog_sales.cs_item_sk = catalog_returns.cr_item_sk
>     and catalog_sales.cs_order_number = catalog_returns.cr_order_number
>   group by cs_item_sk) cs_ui
> ON store_sales.ss_item_sk = cs_ui.cs_item_sk
>   WHERE  
>          i_color in ('maroon','burnished','dim','steel','navajo','chocolate') and
>          i_current_price between 35 and 35 + 10 and
>          i_current_price between 35 + 1 and 35 + 15
> group by d1.d_year,i_item_sk
> ) cs2
> ON cs1.item_sk=cs2.item_sk
> where 
>      cs1.syear = 2000 and
>      cs2.syear = 2000 + 1 and
>      cs2.cnt <= cs1.cnt;
> {code}



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

Mime
View raw message